### 1) Disable scientific notations
options(scipen = 999)
### 2) Install and/or load packages
pacman::p_load(tidyverse,
haven, labelled, sjlabelled, # for working with labeled .sav data
stargazer, ggeffects, dotwhisker, # for generating regression outputs
gt, gtExtras) # for generating tables
### 3) Read the dataset
# Read in the Global Attitudes Survey
Global.df <- read_sav("./Dataset/Global Attitudes Survey (Spring 2021).sav")
# Read in the American Trends Panel
US.df <- read_sav("./Dataset/American Trends Panel (Wave 82).sav")
### 4) (For reference) Create a data dictionary
# A look up table for the original response categories
dictionary_global <- generate_dictionary(Global.df)
dictionary_US <- generate_dictionary(US.df)
# To retrieve the look up table:
# dictionary_US$value_labels or dictionary_global$value_labelsPublic attitudes toward multilateral climate action
R Script
For a cross-country comparison of climate-related attitudes, I combined 2 sets of data from the Pew Research Center:
- Global Attitudes Survey: a global survey across 15 countries, excluding the U.S.
- American Trends Panel: a national survey of U.S. adults
However, the focus of this analysis was limited to the following list of countries, for which climate data were available.
The 6 countries were chosen so as to reflect geographical representation:
- The world’s top 10 biggest polluters:
- U.S.
- Germany
- South Korea
- Countries at the forefront of taking climate action:
- Australia
- Sweden
- The Netherlands
I. Getting started
II. Data wrangling
a. The Global Attitudes Survey
1) Clean column names & filter observations by country
# Remove 1) prefixes 'D_' and 2) suffixes '_2017'
names(Global.df) <- str_remove_all(names(Global.df), "^D_|_2017$")
Global.df <-
Global.df %>%
# Assign descriptive variable names
rename("EDUC_AUSTRALIA_SUPPLEMENTARY" = "EDUC_AUSTRALIA",
"EDUC_AUSTRALIA" = "EDUC_AUSTRALIA_2017B") %>%
# Subset to respondents from
# Australia, Germany, Netherlands, South Korea, and Sweden:
filter(public == 1 | public == 5 | public == 9 | public == 12 | public == 14) %>%
# Instead of displaying numeric values for each national sample,
# assign their descriptive value labels, to indicate respondents' nationality (as strings):
mutate(COUNTRY = get_labels(public)[public])2) Standardize the education variable across countries
# Define a custom function for cleaning up the education-related variables for
# 1) the Dutch, 2) Korean, and 3) the Swede samples:
clean_response1 <- function(x) {
cleaned_NA1 = ifelse(x %in% c(98, 99), NA,# Code 'Don't know' and 'Refused' as NA
ifelse(x %in% c(8, 9, 10), 1, # 'BA degrees or higher' as 1
# If the data pertains to the [FOCAL COUNTRY] AND isn't a missing value, code as 0 ('No BA degree')
# Otherwise (i.e. if the data doesn't actually pertain to the [FOCAL COUNTRY]), code as NA.
ifelse(!is.na(x), 0, NA)))
return(cleaned_NA1)
}
Global_edu_cleaned <-
Global.df %>%
mutate(
# For the Australian sample,
# if the highest year of primary/secondary school completion is 'Don't know' or 'Refused',
# code as NA:
EDUC_AUSTRALIA_TEMPORARY = ifelse(EDUC_AUSTRALIA_SUPPLEMENTARY %in% c(8, 9), NA,
# If the respondent is *NOT* from Australia, also code them as NA. Otherwise, code as 0:
ifelse(COUNTRY != 'Australia', NA, 0)),
# Then, overwrite'EDUC_AUSTRALIA_TEMPORARY' over our measure of 'highest degree obtained',
# *ONLY IF* the latter is missing a value.
# The overwritten values will either be an a) 'NA' or b) '0' (for 'Without a BA degree'):
EDUC_AUSTRALIA_NEW = ifelse(is.na(EDUC_AUSTRALIA), EDUC_AUSTRALIA_TEMPORARY,
ifelse(EDUC_AUSTRALIA %in% c(98, 99), NA, # recode the 'Don't know's and 'Refused's
ifelse(EDUC_AUSTRALIA %in% c(5, 6, 7), 1, 0))), # recode non-missing values as 0 or 1s
# Next, clean the 'EDUC_GERMANY' column:
EDUC_GERMANY_NEW = ifelse(EDUC_GERMANY %in% c(8, 9), NA, # recode the 'Don't know's and 'Refused's
ifelse(EDUC_GERMANY %in% c(5, 6, 7), 1, # recode the 'BA or higher's
# For the remaining observations in this column, code as 0 if the data is from a German sample.
# Otherwise, code as NA (because these observations are from non-German samples):
ifelse(!is.na(EDUC_GERMANY), 0, NA))),
# For 1) the Netherlands, 2) Korea, and 3) Sweden,
# apply the customized function, then append '_NEW' to the original column names:
across(c(EDUC_NETHERLANDS, EDUC_SKOREA, EDUC_SWEDEN), ~ clean_response1(.x), .names = "{col}_NEW")) %>%
# Collapse the recoded responses of all 5 countries into a single variable:
mutate(
EDUCATION = ifelse(COUNTRY == 'Australia', EDUC_AUSTRALIA_NEW,
ifelse(COUNTRY == 'Germany', EDUC_GERMANY_NEW,
ifelse(COUNTRY == 'Netherlands', EDUC_NETHERLANDS_NEW,
ifelse(COUNTRY == 'South Korea', EDUC_SKOREA_NEW, EDUC_SWEDEN_NEW)))))3) Standardize the income variable across countries
# First, code the 'Don't know's and 'Refused' as NA,
# because calculating the median income category requires the removal of NAs
# To do so, define a custom function that converts the missing values of income-related variables:
clean_response2 <- function(x) {
cleaned_NA2 = ifelse(x %in% c(98, 99), NA, x) # keep the non-missing values as-is
return(cleaned_NA2)
}
# Apply the function across all 5 countries:
Global_edu_income_cleaning <-
Global_edu_cleaned %>%
mutate(
across(starts_with("INCOME_") & ends_with(c("AUSTRALIA", "GERMANY", "NETHERLANDS", "SKOREA", "SWEDEN")), ~ clean_response2(.x)),
# Collapse the responses of each country into a single variable:
INCOME_UNSTANDARDIZED = ifelse(COUNTRY == 'Australia', INCOME_AUSTRALIA,
ifelse(COUNTRY == 'Germany', INCOME_GERMANY,
ifelse(COUNTRY == 'Netherlands', INCOME_NETHERLANDS,
ifelse(COUNTRY == 'South Korea', INCOME_SKOREA, INCOME_SWEDEN)))))
# Next, find the median income category for each country
Global_income_median_UNSTANDARDIZED <-
Global_edu_income_cleaning %>%
group_by(COUNTRY) %>%
summarize(median_income = median(INCOME_UNSTANDARDIZED, na.rm = TRUE))
# Create a summary table of median income, for each national sample
Global_income_median_UNSTANDARDIZED %>%
gt() %>%
tab_header(
title = "The median income category of survey respondents",
subtitle = "By country:") %>%
tab_source_note(
source_note = md("**<ins>Note</ins>**: The income categories presented above are *unstandardized*.")) %>%
cols_label(
COUNTRY = "Country/Nationality",
median_income = "Median income category") %>% # rename the columns
cols_align(align = "left") %>% # align columns to the left
gt_theme_538()| The median income category of survey respondents | |
| By country: | |
| Country/Nationality | Median income category |
|---|---|
| Australia | 6 |
| Germany | 6 |
| Netherlands | 5 |
| South Korea | 8 |
| Sweden | 8 |
| Note: The income categories presented above are unstandardized. | |
# Lastly, standardize the income variables so that:
# values above or equal to the *country* median (i.e. High income) = 1 and
# values below the *country* median (i.e. Low income) = 0
Global_edu_income_cleaned <-
Global_edu_income_cleaning %>%
mutate(
INCOME = ifelse(is.na(INCOME_UNSTANDARDIZED), NA, # Code the missing values
# Specify the higher income category among the Australian & German sample:
ifelse((COUNTRY %in% c('Australia', 'Germany') & INCOME_UNSTANDARDIZED >= 6) |
# Higher income category among the Dutch sample:
(COUNTRY == 'Netherlands' & INCOME_UNSTANDARDIZED >= 5) |
# Higher income category among the South Korean sample and the Swedes:
# Then code as 1 if high in income, otherwise code as 0
(COUNTRY %in% c('South Korea', 'Sweden') & INCOME_UNSTANDARDIZED >= 8), 1, 0)))4) Standardize the responses across the Global and the US data
- To perform a logistic regression, convert the dependent variable (
CLIMATE_INTLCOMMUNITY) into a binary variable. Rename it asCLIMATE_CONFIDENCE - Collapse the scale responses and convert the independent variable (
DIVERSITY_GOODBAD) into a binary variable. Rename it asDIVERSITY_GOOD - Code
sex/genderas1= female,0= male/non-female, instead of1s and2s - Collapse the scale responses and recode
political ideologyas1= left,2= moderate,3= right - Classify the respondents’
age(straight age) into age categories
# Create a custom function that converts the missing values in
# our measure of 1) satisfaction with democracy and 2) climate concern:
clean_response3 <- function(x) {
# Code 'Don't know' or 'Refused' as 'NA'. Else, keep it as is.
cleaned_NA3 = ifelse(x %in% c(8, 9), NA, x)
return(cleaned_NA3)
}
# Clean the entire dataset:
Global.df_FINAL <- Global_edu_income_cleaned %>%
mutate(
# Recode 'very confident' or 'somewhat confident' = 1
CLIMATE_CONFIDENCE = ifelse(CLIMATE_INTLCOMMUNITY %in% c(1, 2), 1,
# 'not too confident' or 'not at all confident' = 0
ifelse(CLIMATE_INTLCOMMUNITY %in% c(3, 4), 0, NA)),
DIVERSITY_GOOD = ifelse(DIVERSITY_GOODBAD %in% c(8, 9), NA,
# Recode 'diversity makes a better place to live' = 1
# 'diversity makes a worse place to live' or 'diversity doesn't make much difference' = 0
ifelse(DIVERSITY_GOODBAD == 1, 1, 0)),
# Apply the custom function to convert NAs
across(c(SATISFIED_DEMOCRACY, CLIMATE_CONCERN), ~ clean_response3(.x)),
FEMALE = ifelse(SEX == 2, 1, 0), # Dummy variable for sex/gender
AGE_CATEGORY = case_when(
AGE == 99 ~ NA_real_,
(AGE >= 18 & AGE <= 29) ~ 1,
(AGE >= 30 & AGE <= 49) ~ 2,
(AGE >= 50 & AGE <= 64) ~ 3,
AGE >= 65 ~ 4),
POLITICAL_ID = ifelse(POLITICAL_SCALE2 %in% c(8, 9), NA,
# Code 'extreme left' ~ 'left leaning' as 1
ifelse(POLITICAL_SCALE2 %in% c(0, 1, 2), 1,
# Code 'right leaning' ~ 'extreme right' as 3; and 'center' as 2
ifelse(POLITICAL_SCALE2 %in% c(4, 5, 6), 3, 2)))) %>%
select(
COUNTRY,
CLIMATE_CONFIDENCE, # Dependent variable
SATISFIED_DEMOCRACY, DIVERSITY_GOOD, # Independent variables
CLIMATE_CONCERN, FEMALE, AGE_CATEGORY, POLITICAL_ID, EDUCATION, INCOME) # Controls
# Export the cleaned dataset
write.csv(Global.df_FINAL, file = "./Dataset/Global Attitudes Survey_CLEANED.csv")b. The American Trends Panel
1) Clean column names
# Remove 1) prefixes 'GAP21' and 2) suffixes '_W82' from the column names
names(US.df) <- str_remove_all(names(US.df), "^GAP21|_W82$")2) Standardize the responses across the Global and the US data
US.df_FINAL <-
US.df %>%
mutate(
# Change all 'Refused's to NA
across(c(starts_with("Q"), starts_with("F_")), ~ ifelse(.x == 99, NA, .x)),
# Create a new variable that indicates the nationality of the sample
COUNTRY = 'US',
# Convert to binary variables:
CLIMATE_CONFIDENCE = case_when(
Q28 %in% c(1,2) ~ 1, # Code as "confident"
Q28 %in% c(3,4) ~ 0, # as "not confident"
TRUE ~ Q28),
# 'Diversity makes it a worse place to live' = 0; Otherwise, keep the values as-is
DIVERSITY_GOOD = ifelse(Q20 == 2, 0, Q20),
FEMALE = ifelse(is.na(F_GENDER), NA,
ifelse(F_GENDER == 2, 1, 0)), # female = 1; non-female = 0
EDUCATION = ifelse(F_EDUCCAT %in% c(2,3), 0, F_EDUCCAT), # No BA = 0; Otherwise, keep as-is
# Distinguish between high- vs. low-income
INCOME = ifelse(is.na(F_INC_SDT1), NA,
ifelse(F_INC_SDT1 >= median(F_INC_SDT1, na.rm = TRUE), 1, 0)),
POLITICAL_ID = case_when(
F_IDEO %in% c(4,5) ~ 1, # Code 'Liberal' and 'Very liberal' as 'Left'
F_IDEO == 3 ~ 2, # 'Moderate' as 'Center'
F_IDEO %in% c(1,2) ~ 3, # 'Very conservative' and 'Conservative' as 'Right'
TRUE ~ F_IDEO)) %>%
rename(SATISFIED_DEMOCRACY = Q3, # Assign descriptive variable names
CLIMATE_CONCERN = Q24,
AGE_CATEGORY = F_AGECAT) %>%
select(COUNTRY,
CLIMATE_CONFIDENCE, # Dependent variable
SATISFIED_DEMOCRACY, DIVERSITY_GOOD, # Independent variables
CLIMATE_CONCERN, FEMALE, AGE_CATEGORY, POLITICAL_ID, EDUCATION, INCOME) # Controls
# Export the cleaned dataset
write.csv(US.df_FINAL, file = "./Dataset/American Trends Panel_CLEANED.csv")c. Combine the two data sets
# Merge the Global and the US survey data sets
COMBINED_df <- rbind(Global.df_FINAL, US.df_FINAL)
# Export the compiled version of the dataset
write.csv(COMBINED_df, file = "./Dataset/Climate Attitudes Survey_FINAL.csv")
# Split the data frame according to each country
# (In order to fit 6 different regression models)
national_sample <- split(COMBINED_df, f = COMBINED_df$COUNTRY) III. Data analysis
1) Print the summary statistics
# Summary statistics of the entire sample (across 6 countries)
summary(COMBINED_df)
# Summary statistics of each national sample
countries <- c("Australia", "South Korea", "Germany", "Netherlands", "Sweden", "US")
country_summary <- function(data, countries) {
summaries <- map(countries, ~ summary(data[[.x]]))
names(summaries) <- countries
return(summaries)
}
country_summary(national_sample, countries)2) Create 2 contingency tables for each country
# one for attitudes toward diversity (X1) and the other for satisfaction with democracy (X2)
# For examining the distribution of 0s and 1s of Y, across varying levels of X
variables <- c("SATISFIED_DEMOCRACY", "DIVERSITY_GOOD")
generate_cross_tabs <- function(country, X) {
xtabs(~ CLIMATE_CONFIDENCE + get(X), data = national_sample[[country]])
}
for (country in countries) {
cat("Country:", country, "\n")
# Generate cross-tabulations
# For attitudes toward diversity (X1):
cat("For DIVERSITY_GOOD:\n")
print(map(country, ~generate_cross_tabs(.x, "DIVERSITY_GOOD")))
# For satisfaction with democracy (X2):
cat("For SATISFIED_DEMOCRACY:\n")
print(map(country, ~generate_cross_tabs(.x, "SATISFIED_DEMOCRACY")))
}3) Fit a multiple logistic regression model
# Define the equation for the logistic regression model
logit_equation <-
CLIMATE_CONFIDENCE ~ # Y variable
DIVERSITY_GOOD + SATISFIED_DEMOCRACY + # X variables
CLIMATE_CONCERN + FEMALE + AGE_CATEGORY + EDUCATION + POLITICAL_ID + INCOME # Control variables
# For each country, fit the model
models <- list()
for (country in countries) {
data <- national_sample[[country]]
glm_model <- glm(logit_equation, data = data, family = binomial("logit"))
models[[country]] <- glm_model
}4) Visualize the results of the regression analysis
# Create a regression table
regression_table <- stargazer(
models,
title = "The effect of diversity and democracy perceptions on perceived efficacy of multilateral climate action",
column.labels = toupper(countries),
align = TRUE,
type = "text")
The effect of diversity and democracy perceptions on perceived efficacy of multilateral climate action
====================================================================================
Dependent variable:
----------------------------------------------------------------
CLIMATE_CONFIDENCE
AUSTRALIA SOUTH KOREA GERMANY NETHERLANDS SWEDEN US
(1) (2) (3) (4) (5) (6)
------------------------------------------------------------------------------------
DIVERSITY_GOOD 0.344 0.416** 0.463** 0.424*** 0.560*** 0.568***
(0.246) (0.163) (0.186) (0.162) (0.198) (0.188)
SATISFIED_DEMOCRACY -0.288*** -0.184* -0.393*** -0.655*** -0.438*** -0.720***
(0.090) (0.097) (0.079) (0.087) (0.098) (0.064)
CLIMATE_CONCERN -0.410*** -0.472*** -0.082 -0.172** -0.128 -0.795***
(0.083) (0.107) (0.079) (0.084) (0.081) (0.059)
FEMALE 0.308** 0.305* 0.181 0.097 -0.141 0.288***
(0.151) (0.182) (0.150) (0.151) (0.144) (0.099)
AGE_CATEGORY -0.021 0.113 0.230*** 0.123* -0.086 -0.056
(0.076) (0.091) (0.074) (0.071) (0.072) (0.050)
EDUCATION -0.355** 0.156 -0.184 -0.625*** -0.134 -0.095
(0.162) (0.169) (0.163) (0.163) (0.157) (0.108)
POLITICAL_ID -0.074 -0.083 0.128 0.170* 0.081 -0.346***
(0.110) (0.117) (0.104) (0.098) (0.095) (0.073)
INCOME 0.059 0.055 -0.406*** -0.050 -0.481*** -0.047
(0.166) (0.173) (0.155) (0.165) (0.151) (0.107)
Constant 1.236** 1.588*** 0.225 1.483*** 0.611 3.770***
(0.511) (0.464) (0.432) (0.441) (0.449) (0.335)
------------------------------------------------------------------------------------
Observations 782 807 834 860 960 2,382
Log Likelihood -512.243 -472.542 -541.840 -538.488 -605.132 -1,273.024
Akaike Inf. Crit. 1,042.486 963.084 1,101.679 1,094.975 1,228.264 2,564.047
====================================================================================
Note: *p<0.1; **p<0.05; ***p<0.01# Plot the coefficients and their 95% CI
# Plot the log odds
dwplot(models) %>%
relabel_predictors(
c(DIVERSITY_GOOD = "DIVERSITY IS \n BETTER",
SATISFIED_DEMOCRACY = "SATISFACTION \n WITH DEMOCRACY \n AT HOME",
CLIMATE_CONCERN = "CONCERNED ABOUT \n CLIMATE CHANGE",
FEMALE = "FEMALE",
AGE_CATEGORY = "AGE",
EDUCATION = "EDUCATION",
POLITICAL_ID = "POLITICAL \n IDEOLOGY",
INCOME = "INCOME")) +
theme_gray() +
geom_vline(xintercept = 0, linetype = 2) + # Add a dotted line at x = 0
labs(
title = "Who tends to be more optimistic of \n tackling climate action at the international level?",
subtitle = "Positive attitudes toward social diversity and low satisfaction with the functioning of democracy
within one's country are good predictors of one's confidence in multilateral solutions",
x = "Coefficient Estimates \n (in log odds)",
caption = "Source: The Pew Research Center. \n Note: Bars represent 95% Confidence Intervals.") +
scale_color_discrete(name = "Countries",
labels = countries) +
theme(plot.title = element_text(hjust = 0.5, face = "bold"),
plot.subtitle = element_text(hjust = 0.5),
plot.caption = element_text(hjust = 1, size = 9),
text = element_text(size = 10.5))
5) Calculate the predicted probabilities
# For each country, compute the predicted probability that an individual would be
# optimistic of the impacts of global climate action, across different levels of
# a) diversity perception and b) satisfaction with democracy
# Hold other (independent and control) variables constant at the median
# In order to draw comparisons across multiple countries,
# set the reference point to the global median (i.e. the median of all 6 countries),
# NOT the national median
# Create a list of median values, which we will pass as the parameters of ggpredict()
# Set the reference to the *global* median:
median_values <- c(
SATISFIED_DEMOCRACY = 2,
DIVERSITY_GOOD = 1,
CLIMATE_CONCERN = 2,
FEMALE = 0,
AGE_CATEGORY = 3,
POLITICAL_ID = 2,
EDUCATION = 0,
INCOME = 1)
# Store the predicted probabilities for a given model and term
generate_predicted_probs <- function(model, X_var) {
ggpredict(model, terms = NULL, condition = median_values, pretty = TRUE)
}6) Plot the predicted probabilities
predicted_probs_list <- map(models, ~ generate_predicted_probs(.x))
# Predicted probabilities for diversity perceptions
# 1. Create a predicted probabilities table that compares across individuals' social diversity beliefs
(predicted_probs_DIVERSITY <-
predicted_probs_list %>%
# Return it as a dataframe
map_dfr(~ .x$DIVERSITY_GOOD, .id = "Country") %>%
# Give a descriptive label to the 'x' column, which indicates different levels of social diversity beliefs:
rename(DIVERSITY_GOOD = x) %>%
# 2. Plot the predicted probabilities and confidence intervals
ggplot(aes(
x = factor(DIVERSITY_GOOD),
y = predicted,
color = Country)) +
geom_point() +
geom_errorbar(aes(
ymax = conf.high, ymin = conf.low),
alpha = 0.25,
width = 0.5,
color = "black") +
scale_y_continuous(
limits = c(0, 1),
n.breaks = 6) +
scale_color_discrete(
breaks = c(0:5),
labels = countries) +
labs(
title = "The predicted probabilities of the level of confidence \n in our ability to tackle global climate change via multilateralism",
x = "Diversity makes the world a better place",
y = "Predicted probability",
subtitle = "A cross-country comparison along diversity perceptions",
caption = "Source: The Pew Research Center. \n Note: Bars represent 95% Confidence Intervals.
The graph shows the predicted probabilities while holding all other values constant at their median.") +
theme(
plot.title = element_text(hjust = 0.5, face = "bold"),
plot.subtitle = element_text(hjust = 0.5),
plot.caption = element_text(hjust = 1, size = 9),
text = element_text(size = 10.5)) +
facet_wrap(~ Country, nrow = 1))
# Predicted probabilities for democracy perceptions
# 1. Create a predicted probabilities table that compares across different levels of satisfaction with democracy
(predicted_probs_DEMOCRACY <-
predicted_probs_list %>%
# Return it as a dataframe
map_dfr(~ .x$SATISFIED_DEMOCRACY, .id = "Country") %>%
# Give a descriptive label to the 'x' column, which refers to different levels of satisfaction with democracy:
rename(SATISFIED_DEMOCRACY = x) %>%
# 2. Plot the predicted probabilities and confidence intervals
ggplot(aes(
x = SATISFIED_DEMOCRACY,
y = predicted,
linetype = Country)) +
geom_line() +
geom_ribbon(aes(
ymax = conf.high, ymin = conf.low),
alpha = 0.25) +
scale_y_continuous(
limits = c(0, 1),
n.breaks = 6) +
scale_linetype_discrete(
breaks = c(0:5),
labels = countries) +
labs(
title = "The predicted probabilities of the level of confidence \n in our ability to tackle global climate change via multilateralism",
x = "Satisfaction with the way democracy functions at home",
y = "Predicted Probability",
subtitle = "A cross-country comparison along democracy perceptions",
caption = "Source: The Pew Research Center. \n Note: Bars represent 95% Confidence Intervals. \n The graph shows the predicted probabilities while holding all other values constant at their median.") +
theme(
plot.title = element_text(hjust = 0.5, face = "bold"),
plot.subtitle = element_text(hjust = 0.5),
plot.caption = element_text(hjust = 1, size = 9),
text = element_text(size = 10.5)) +
facet_wrap(~ Country, nrow = 1))